Re: [Xen-devel] [PATCH V4 03/15] create arch functions to get and process EFI memory map.

[Ian Campbell <Ian.Campbell@xxxxxxxxxx>]

On Fri, 2014-09-12 at 10:56 +0100, Jan Beulich wrote:
> >>> On 12.09.14 at 11:45, <Ian.Campbell@xxxxxxxxxx> wrote:
> > On Fri, 2014-09-12 at 08:07 +0100, Jan Beulich wrote:
> >> >>> On 11.09.14 at 19:40, <roy.franz@xxxxxxxxxx> wrote:
> >> > On Thu, Sep 11, 2014 at 7:11 AM, Jan Beulich <JBeulich@xxxxxxxx> wrote:
> >> >>>>> On 10.09.14 at 02:51, <roy.franz@xxxxxxxxxx> wrote:
> >> >>> @@ -1171,67 +1169,12 @@ efi_start(EFI_HANDLE ImageHandle, 
> >> >>> EFI_SYSTEM_TABLE 
> > *SystemTable)
> >> >>>          }
> >> >>>      }
> >> >>>
> >> >>> -    efi_bs->GetMemoryMap(&efi_memmap_size, NULL, &map_key,
> >> >>> -                         &efi_mdesc_size, &mdesc_ver);
> >> >>> -    mbi.mem_upper -= efi_memmap_size;
> >> >>> -    mbi.mem_upper &= -__alignof__(EFI_MEMORY_DESCRIPTOR);
> >> >>> -    if ( mbi.mem_upper < xen_phys_start )
> >> >>> -        blexit(L"Out of static memory");
> >> >>> -    efi_memmap = (void *)(long)mbi.mem_upper;
> >> >>> -    status = efi_bs->GetMemoryMap(&efi_memmap_size, efi_memmap, 
> >> >>> &map_key,
> >> >>> -                                  &efi_mdesc_size, &mdesc_ver);
> >> >>> -    if ( EFI_ERROR(status) )
> >> >>> -        PrintErrMesg(L"Cannot obtain memory map", status);
> >> >>> +    efi_arch_get_memory_map(&mmap_size, &mmap, &mmap_key,
> >> >>> +                                  &mmap_desc_size, &mmap_desc_ver);
> >> >>
> >> >> The only arch-specific bit here is where to put the map.
> >> > Yes, but the ARM method of allocating the space can make the map
> >> > bigger.
> > 
> > So something which ARM does can make the result of efi_bs->GetMemoryMap
> > longer? Or are we talking about the bootinfo.mem array?
> 
> Not having looked at the code, I suppose the issue is with the
> sequence of
> - get memory map size
> - alloc memory for the map
> - get memory map
> where the middle step may alter the memory map size. But with any
> kind of sane allocator it ought to be reasonable to expect it to not
> grow by more than two entries (if the allocator chooses to take the
> middle of an existing free block), which could be accounted for up
> front.

I see, thanks.

Does x86 avoid this by using a static buffer or something? Or by the
"account for it up front" trick which you mention?

Ian.


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel